XGBOOST - objective function to optimise the recall of two minor classes

I currently have a 3-class problem where the 1st class completely dominates the other two classes, and I am trying to boost the recall in the 2nd and 3rd class.

I was wondering if there was an objective function that focusses on recall rather than accuracy, and maybe just on the recall of the 2nd and 3rd class.
Additionally what eval_metric would one choose?