The implementation you are trying to achieve is the same as a "Transfer Learning system". The right way to do this would be load the weights and retrain the model using the new dataset. In order to ensure that the original model is not majorly affected, kindly choose an appropriate "Learning Rate" (generally, a lower learning rate to the prior training)
"Transfer Learning" is achievable with the "Deep Learning Toolbox", kindly refer to the link below:
