if the no: of outliers is very less then they can be deleted from the dataset , if they constitute to majority of the data then we can replace them using mean or median for numerical data and mode for categorical dataset.

A Very first step is to detect the outliers and then to remove those.there are many methods to detect the outliers.our main task is to find real outliers sometimes we cosider some points to be an outlier and remove it.
if data set is too large then will not affect the model if  outliers are removed.but if data set is small then to remove outlier can be dangerous.