Keras - Connect one layer to another in backwards direction, how?

Question

So, I have this basic neural network architecture, where I is input and A,B,C are hiden layers and O is the input like this:

I --> A --> B --> C --> O

So far, this architecture is not an RNN, its a basic feed-forward NN architecture.

But now, here comes the recurrency:

C is connected to A (in backward direction)

hence making it an RNN.

How can i connect the C layer to the A layer in Keras?

Edit :

What i want is the code for a custom layer that takes input from both input layer and another layer (let's call it C)