GridSearchCV and cross_val_score recently adopted support for a scoring object instead of score_func and loss_func (#1381). sklearn.feature_selection.RFECV also accepts loss_func or alternatively uses estimator.score, and should instead support the same interface as GridSearchCV.