Add a LSTM layer to a sequential Keras model and save it accoring to a custom function

Hi, I’m learning about LSTMs and I am working on a simple project using Keras. Do you think you could help me: 1) Add a LSTM layer to my sequential model; 2) Show me how to save the model when it scores higher on a custom scoring function (I want to save it only when the training accuracy is higher than the validation accuracy and both are higher than the previous best combination of testing and validation accuracy)