Function fitting Neural Network from scratch (with at least two hidden layer)

I want to build a function fitting feedforward neural network from scratch without any neural network toolbox help. The performance equation obviously be MSE with backpropagation algorithm. Can anyone help me ?

[Greg Heath]

Why in the world do you need more than a single hidden layer???

Greg

[Shritej Chavan]

Because I think the input-output fitting is really difficult in my case.