Managed Wrapper for CUDA Classes

I have written a simple piece of code that uses CUDA. I am using CUDA_VS_WIZARD from here to compile my code from Visual Studio 2008. I know the code works, I can compile it as an executable and run it. I changed the settings to build the CUDA code into a lib (I changed this setting both in the standard C++ section as well as setting the build type of the CUDA code to lib). This will compile without any complaints. The class that I want to wrap is in a normal C++ header file (.h extension) and has its classes defined across several files. All of the classes that I try to use from within the wrapper cause an unresolved external error.

I have written managed wrappers before without any problems. Has anybody written a managed wrapper for CUDA functions before? Is there something specific to the CUDA compiler that I need to know about for writing managed wrappers or I am being an idiot and missing something elsewhere?

Thanks.