Why can't you use cuda / opencl to force updates on specific video memory?

I guess trying to mix directx / opengl and cuda / opencl would be a pain. One is at a completely different abstraction layer than the other.