Keras is based on deep learning that allows you to create any kind of neural network. Keras support neural networks that are very simple as well as complex. To better understand the working of the Keras framework, you can refer to the following Architecture of Keras:

There are three main API layers of Keras architecture that are described as follows:

• Model API: There are two types of Keras models, i.e. Sequential and Functional. The sequential model is a linear composition of all layers of the Keras model. It is easy to implement and allows you to show all available neural networks. The Functional API model is simply used to create complex deep-learning models. 
• Core Modules API: There is a huge number of built-in neural network functions that contribute to creating the Keras model and its layers. These functions are based on some models, such as the Optimizer module, Regularizers, Activations module, and loss module. 
• Layer API: There are several built-in layers in Keras that are used to create a complex neural network. Some important layers include Core layers, pooling layers, convolution layers, and recurrent layers.