What's the best way to deploy a complicated machine learning model into a simple environment that has no libraries?

For example, I can develop a model (say a random forest) in R but then have to deploy it where I can't use R and have to write all the code myself.

A possible way is to use R to generate all possible predictions and then fit or even over fit a linear regression model to my predictions aiming for 100% R squared. This could be difficult to achieve with linear models.

Any better methods?