[bri3d]

As you can see from the documented reverse-engineering of the hardware here:

http://linux-sunxi.org/VE_Register_guide

the device is generally single-function for each codec, with the decoding pipeline implemented in actual hardware. Unless you were able to express your kernel as a data transform used in a common video codec and express your I/O data as pixels, you're unlikely to accelerate a general purpose task this way.