Feed input to a keras model only to be accessed by a custom loss function

Question

Is there a way to pass a feature to a keras model as an input only to be accessed by a custom loss function without affecting the model as an input feature? I only need the