Combining features from multiple models and optimising features

I have multiple models predicting an outcome (continuous) and I want to take action to optimize the best values of these features to make a decision.

Consider a regression model, y1 = m1x1 + m2x2 + m3x3 + k

and another mode preponsity model y2 = P[1 | x5, x6, x3] = 1 / (1 + exp(-m5x5 - m6x2 - m7x3))

I would want to maximize y3 say y3 = f(y1, y2) by finding the optimal values of the features x1, x2, x3, ... x5

Questions -

1. Any ideas to formulate a problem that can help me solve the above?
2. If I have additional features that are not used in the above two models, I want to use them to find the right points to maximize y3.