It should be possible. However, does that make sense?
CFA arrays are 16-bit integer data, so I don't see how interpolating them to 64-bit floating point numbers could be of any benefit. Currently, the Debayer process generates a 32-bit floating point image, which is much more than sufficient IMO.
Or are you planning to do something more creative here ?