When to enter a confounder into a hierarchical regression model?

Hello all, two questions!:

1. I can't find any resources on this but just wanted to check - am I right to think that to assess the effect of controlling for a confounding variable, you would enter it into a hierarchical regression model at the last step? In other words, you would look at the predictive power of the predictors you are interested in first, and then look to see if that is maintained AFTER controlling for the confounder?

2. I have a bunch of theoretically-selected predictors (3 in total) for a number of DVs. As I see it, I have the options of a) contructing individual simple regression models for each predictor and outcome variable b) constructing a multiple regression model for each DV with all three predictors entered or c) as with b, and then refining the model to include only the significant predictors. Is this correct, and is there any reason to choose one over the others?

Thank you!