metric distances estimation

A metric allows to estimate similarity if there is one that uses important features as arguments.  How would you construct a metric that gave a weight to each feature according to their importance?  Would it make sense to normalize the values of the features and how would you do it? How would you find the importance of the features relative to one another?