1) If you use the same model for each of the folds it will not help, the last fold will be overfitted.

2) If you use different models for each fold, you will have k models that you can put into an ensemble predictor, but that goes out of the scope of k-fold cross-validation, which again, is just an evaluation strategy.