Suggestion, feel free to ignore:
when adding smaller units use boid theory to have them swarm. If you wish to go the extra mile, you can use Q-Learning for medium units so that the military gets smarter the more they fight you. As for mechanics, I suggest using Singular Point as an example of how to do upgrades. In order to keep away the pain of programming every single enemy, have the AI fight an AI version of the kaiju with Q-Learning and save the weights for distribution. This way, they'll seem smarter. You can find papers for free on academia.edu