Custom Neural Network Architectures

Hi all,

I’m new to this forum and deep learning so I would be grateful if someone could point a few things out.
I have a programming and software developing background, yet this transition is a little difficult.

I want to create, train and use custom NN architectures with CUDA but i’m a little lost with all the available framework options.
The training data is completely custom as well and as far as I have seen, available frameworks specialize in vision, natural language etc.
Can anyone point out what the best C-compatible framework is for building custom NN’s and training them on custom data (using existing algorithms like Backpropagation, Adadelta, Momentum, etc.)?
Also, any info on how to get started with said framework is greatly appreciated!

Thanks in advance,
Alex