diff --git a/source/nvidia/cuda/nvidia-cuda-context.cpp b/source/nvidia/cuda/nvidia-cuda-context.cpp
index e738c93c..2cb250c3 100644
--- a/source/nvidia/cuda/nvidia-cuda-context.cpp
+++ b/source/nvidia/cuda/nvidia-cuda-context.cpp
@@ -31,13 +31,7 @@
 #endif
 #endif
 
-nvidia::cuda::context::context(std::shared_ptr<::nvidia::cuda::cuda> cuda)
-	: _cuda(cuda), _ctx(), _has_device(false), _device()
-{
-	if (!cuda)
-		throw std::invalid_argument("cuda");
-}
-
+nvidia::cuda::context::context() : _cuda(::nvidia::cuda::cuda::get()), _ctx(), _has_device(false), _device() {}
 nvidia::cuda::context::~context()
 {
 	if (_has_device) {
@@ -47,7 +41,7 @@ nvidia::cuda::context::~context()
 }
 
 #ifdef WIN32
-nvidia::cuda::context::context(std::shared_ptr<::nvidia::cuda::cuda> cuda, ID3D11Device* device) : context(cuda)
+nvidia::cuda::context::context(ID3D11Device* device) : context()
 {
 	using namespace nvidia::cuda;
 
diff --git a/source/nvidia/cuda/nvidia-cuda-context.hpp b/source/nvidia/cuda/nvidia-cuda-context.hpp
index 1ae042df..de9b42db 100644
--- a/source/nvidia/cuda/nvidia-cuda-context.hpp
+++ b/source/nvidia/cuda/nvidia-cuda-context.hpp
@@ -31,13 +31,13 @@ namespace nvidia::cuda {
 		::nvidia::cuda::device_t _device;
 
 		private:
-		context(std::shared_ptr<::nvidia::cuda::cuda> cuda);
+		context();
 
 		public:
 		~context();
 
 #ifdef WIN32
-		context(std::shared_ptr<::nvidia::cuda::cuda> cuda, ID3D11Device* device);
+		context(ID3D11Device* device);
 #endif
 
 		::nvidia::cuda::context_t get();
diff --git a/source/nvidia/cuda/nvidia-cuda-gs-texture.cpp b/source/nvidia/cuda/nvidia-cuda-gs-texture.cpp
index a9aa0f74..19e51c22 100644
--- a/source/nvidia/cuda/nvidia-cuda-gs-texture.cpp
+++ b/source/nvidia/cuda/nvidia-cuda-gs-texture.cpp
@@ -20,13 +20,11 @@
 #include "nvidia-cuda-gs-texture.hpp"
 #include "obs/gs/gs-helper.hpp"
 
-nvidia::cuda::gstexture::gstexture(std::shared_ptr<nvidia::cuda::cuda> cuda, std::shared_ptr<gs::texture> texture)
-	: _cuda(cuda), _texture(texture), _resource(), _is_mapped(false), _pointer()
+nvidia::cuda::gstexture::gstexture(std::shared_ptr<gs::texture> texture)
+	: _cuda(::nvidia::cuda::cuda::get()), _texture(texture), _resource(), _is_mapped(false), _pointer()
 {
 	if (!texture)
 		throw std::invalid_argument("texture");
-	if (!cuda)
-		throw std::invalid_argument("cuda");
 
 	gs::context gctx;
 	int         dev_type = gs_get_device_type();
diff --git a/source/nvidia/cuda/nvidia-cuda-gs-texture.hpp b/source/nvidia/cuda/nvidia-cuda-gs-texture.hpp
index 26e8f90c..3e63bad1 100644
--- a/source/nvidia/cuda/nvidia-cuda-gs-texture.hpp
+++ b/source/nvidia/cuda/nvidia-cuda-gs-texture.hpp
@@ -35,7 +35,7 @@ namespace nvidia::cuda {
 		std::shared_ptr<nvidia::cuda::stream> _stream;
 
 		public:
-		gstexture(std::shared_ptr<nvidia::cuda::cuda> cuda, std::shared_ptr<gs::texture> texture);
+		gstexture(std::shared_ptr<gs::texture> texture);
 		~gstexture();
 
 		array_t map(std::shared_ptr<nvidia::cuda::stream> stream);
diff --git a/source/nvidia/cuda/nvidia-cuda-memory.cpp b/source/nvidia/cuda/nvidia-cuda-memory.cpp
index 76036e6a..dfe9f344 100644
--- a/source/nvidia/cuda/nvidia-cuda-memory.cpp
+++ b/source/nvidia/cuda/nvidia-cuda-memory.cpp
@@ -20,10 +20,9 @@
 #include "nvidia-cuda-memory.hpp"
 #include <stdexcept>
 
-nvidia::cuda::memory::memory(std::shared_ptr<::nvidia::cuda::cuda> cuda, std::size_t size)
-	: _cuda(cuda), _pointer(), _size(size)
+nvidia::cuda::memory::memory(size_t size) : _cuda(::nvidia::cuda::cuda::get()), _pointer(), _size(size)
 {
-	::nvidia::cuda::result res = _cuda->cuMemAlloc(&_pointer, size);
+	::nvidia::cuda::result res = _cuda->cuMemAlloc(&_pointer, _size);
 	switch (res) {
 	case ::nvidia::cuda::result::SUCCESS:
 		break;
@@ -36,7 +35,6 @@ nvidia::cuda::memory::~memory()
 {
 	_cuda->cuMemFree(_pointer);
 }
-
 nvidia::cuda::device_ptr_t nvidia::cuda::memory::get()
 {
 	return _pointer;
diff --git a/source/nvidia/cuda/nvidia-cuda-memory.hpp b/source/nvidia/cuda/nvidia-cuda-memory.hpp
index 315aefba..ee40ee4f 100644
--- a/source/nvidia/cuda/nvidia-cuda-memory.hpp
+++ b/source/nvidia/cuda/nvidia-cuda-memory.hpp
@@ -26,10 +26,10 @@ namespace nvidia::cuda {
 	class memory {
 		std::shared_ptr<::nvidia::cuda::cuda> _cuda;
 		device_ptr_t                          _pointer;
-		std::size_t                           _size;
+		size_t                                _size;
 
 		public:
-		memory(std::shared_ptr<::nvidia::cuda::cuda> cuda, std::size_t size);
+		memory(size_t size);
 		~memory();
 
 		device_ptr_t get();
diff --git a/source/nvidia/cuda/nvidia-cuda-stream.cpp b/source/nvidia/cuda/nvidia-cuda-stream.cpp
index 455758b0..b86f40c8 100644
--- a/source/nvidia/cuda/nvidia-cuda-stream.cpp
+++ b/source/nvidia/cuda/nvidia-cuda-stream.cpp
@@ -20,9 +20,7 @@
 #include "nvidia-cuda-stream.hpp"
 #include <stdexcept>
 
-nvidia::cuda::stream::stream(std::shared_ptr<::nvidia::cuda::cuda> cuda, ::nvidia::cuda::stream_flags flags,
-							 int32_t priority)
-	: _cuda(cuda)
+nvidia::cuda::stream::stream(::nvidia::cuda::stream_flags flags, int32_t priority) : _cuda(::nvidia::cuda::cuda::get())
 {
 	nvidia::cuda::result res;
 	if (priority == 0) {
diff --git a/source/nvidia/cuda/nvidia-cuda-stream.hpp b/source/nvidia/cuda/nvidia-cuda-stream.hpp
index a17bbd53..ecbb8599 100644
--- a/source/nvidia/cuda/nvidia-cuda-stream.hpp
+++ b/source/nvidia/cuda/nvidia-cuda-stream.hpp
@@ -27,9 +27,8 @@ namespace nvidia::cuda {
 		::nvidia::cuda::stream_t              _stream;
 
 		public:
-		stream(std::shared_ptr<::nvidia::cuda::cuda> cuda,
-			   ::nvidia::cuda::stream_flags flags = ::nvidia::cuda::stream_flags::DEFAULT, int32_t priority = 0);
 		~stream();
+		stream(::nvidia::cuda::stream_flags flags = ::nvidia::cuda::stream_flags::DEFAULT, int32_t priority = 0);
 
 		::nvidia::cuda::stream_t get();
 	};
diff --git a/source/nvidia/cuda/nvidia-cuda.cpp b/source/nvidia/cuda/nvidia-cuda.cpp
index 80aa07ce..a0edd92f 100644
--- a/source/nvidia/cuda/nvidia-cuda.cpp
+++ b/source/nvidia/cuda/nvidia-cuda.cpp
@@ -18,8 +18,22 @@
  */
 
 #include "nvidia-cuda.hpp"
-#include "common.hpp"
 #include <mutex>
+#include "util/util-logging.hpp"
+
+#ifdef _DEBUG
+#define ST_PREFIX "<%s> "
+#define D_LOG_ERROR(x, ...) P_LOG_ERROR(ST_PREFIX##x, __FUNCTION_SIG__, __VA_ARGS__)
+#define D_LOG_WARNING(x, ...) P_LOG_WARN(ST_PREFIX##x, __FUNCTION_SIG__, __VA_ARGS__)
+#define D_LOG_INFO(x, ...) P_LOG_INFO(ST_PREFIX##x, __FUNCTION_SIG__, __VA_ARGS__)
+#define D_LOG_DEBUG(x, ...) P_LOG_DEBUG(ST_PREFIX##x, __FUNCTION_SIG__, __VA_ARGS__)
+#else
+#define ST_PREFIX "<nvidia::cuda::cuda> "
+#define D_LOG_ERROR(...) P_LOG_ERROR(ST_PREFIX __VA_ARGS__)
+#define D_LOG_WARNING(...) P_LOG_WARN(ST_PREFIX __VA_ARGS__)
+#define D_LOG_INFO(...) P_LOG_INFO(ST_PREFIX __VA_ARGS__)
+#define D_LOG_DEBUG(...) P_LOG_DEBUG(ST_PREFIX __VA_ARGS__)
+#endif
 
 #if defined(_WIN32) || defined(_WIN64)
 #define CUDA_NAME "nvcuda.dll"
@@ -29,28 +43,33 @@
 
 #define CUDA_LOAD_SYMBOL(NAME)                                                            \
 	{                                                                                     \
-		NAME = reinterpret_cast<decltype(NAME)>(os_dlsym(_library, #NAME));                    \
+		NAME = reinterpret_cast<decltype(NAME)>(_library->load_symbol(#NAME));            \
 		if (!NAME)                                                                        \
 			throw std::runtime_error("Failed to load '" #NAME "' from '" CUDA_NAME "'."); \
 	}
 #define CUDA_LOAD_SYMBOL_V2(NAME)                                                         \
 	{                                                                                     \
-		NAME = reinterpret_cast<decltype(NAME)>(os_dlsym(_library, #NAME "_v2"));              \
+		NAME = reinterpret_cast<decltype(NAME)>(_library->load_symbol(#NAME "_v2"));      \
 		if (!NAME)                                                                        \
 			throw std::runtime_error("Failed to load '" #NAME "' from '" CUDA_NAME "'."); \
 	}
 #define CUDA_LOAD_SYMBOL_EX(NAME, OVERRIDE)                                               \
 	{                                                                                     \
-		NAME = reinterpret_cast<decltype(NAME)>(os_dlsym(_library, #OVERRIDE));                \
+		NAME = reinterpret_cast<decltype(NAME)>(_library->load_symbol(#OVERRIDE));        \
 		if (!NAME)                                                                        \
 			throw std::runtime_error("Failed to load '" #NAME "' from '" CUDA_NAME "'."); \
 	}
 
-nvidia::cuda::cuda::cuda()
+nvidia::cuda::cuda::~cuda()
 {
-	_library = os_dlopen(CUDA_NAME);
-	if (!_library)
-		throw std::runtime_error("Failed to load '" CUDA_NAME "'.");
+	D_LOG_DEBUG("Finalizing... (Addr: 0x%" PRIuPTR ")", this);
+}
+
+nvidia::cuda::cuda::cuda() : _library()
+{
+	D_LOG_DEBUG("Initialization... (Addr: 0x%" PRIuPTR ")", this);
+
+	_library = util::library::load(std::string_view(CUDA_NAME));
 
 	// Initialization
 	CUDA_LOAD_SYMBOL(cuInit);
@@ -117,19 +136,16 @@ nvidia::cuda::cuda::cuda()
 	cuInit(0);
 }
 
-nvidia::cuda::cuda::~cuda()
-{
-	os_dlclose(_library);
-}
-
 std::shared_ptr<nvidia::cuda::cuda> nvidia::cuda::cuda::get()
 {
-	static std::shared_ptr<nvidia::cuda::cuda> instance;
-	static std::mutex                          lock;
+	static std::weak_ptr<nvidia::cuda::cuda> instance;
+	static std::mutex                        lock;
 
 	std::unique_lock<std::mutex> ul(lock);
-	if (!instance) {
-		instance = std::make_shared<nvidia::cuda::cuda>();
+	if (instance.expired()) {
+		auto hard_instance = std::make_shared<nvidia::cuda::cuda>();
+		instance           = hard_instance;
+		return hard_instance;
 	}
-	return instance;
+	return instance.lock();
 }
diff --git a/source/nvidia/cuda/nvidia-cuda.hpp b/source/nvidia/cuda/nvidia-cuda.hpp
index 45e9eb73..90b9da31 100644
--- a/source/nvidia/cuda/nvidia-cuda.hpp
+++ b/source/nvidia/cuda/nvidia-cuda.hpp
@@ -18,10 +18,9 @@
  */
 
 #pragma once
-#include "common.hpp"
 #include <cstddef>
-#include <functional>
-#include <memory>
+#include "util/util-bitmask.hpp"
+#include "util/util-library.hpp"
 
 #ifdef WIN32
 #pragma warning(push)
@@ -128,8 +127,7 @@ namespace nvidia::cuda {
 	};
 
 	class cuda {
-		private:
-		void* _library;
+		std::shared_ptr<util::library> _library;
 
 		public:
 		cuda();
@@ -395,7 +393,7 @@ namespace nvidia::cuda {
 							 ID3D11Resource* d3dresource, uint32_t flags);
 #endif
 		public:
-		static std::shared_ptr<cuda> get();
+		static std::shared_ptr<::nvidia::cuda::cuda> get();
 	};
 } // namespace nvidia::cuda