embeding device code in executable with cuda driver model

I’m looking to use the driver model with my kernel to remove the dependency on the cuda runtime for a library I’m distributing, but I don’t want to load the ptx from file at runtime to avoid issues with directory structure (same thing I’m trying to avoid with the runtime, not to distribute files other than the dll). Is there a way to embed this code inside the dll and load it from the dll resource rather than from file?

I’m running under windows for this and using visual studio 2008 at the moment.