Your indexes will be smaller and more efficient if they only index the subset of documents you are potentially querying. @prasad and @keshav_m can point you to dzone articles about this.
See N1QL Resources: Workshop, Articles and more