Correct approach to caching a computationally expensive object generated within function

Question

Background

In the following scenario, I\'m looking at the following workflow:

1. Top-level runner function, in the example below running_function