custom autoencoders by using feedforward nets

Hello, as the build in trainAutoencoder function seems not the feature multi gpu training and validation test, I want to use custom autoencoders by using feedforward nets. Also by doing so, it should be possible to use AE for regression.

As the msesparse performance function works with feedworwardnet I don't face any problems so far. But to be safe, are there any know issues doing this?