Genome sequencing generates large volumes of data and hence requires increasingly higher computational resources. The growing data problem is even more acute in metagenomics applications, where data from an environmental sample include many organisms instead of just one for the common single organism sequencing. Traditional taxonomic classification and clustering approaches and platforms - while designed to be computationally efficient - are not capable of incrementally updating a previously trained system when new data arrive, which then requires complete re-training with the augmented (old plus new) data. Such complete retraining is inefficient and leads to poor utilization of computational resources. An ability to update a classification system with only new data offers a much lower run-time as new data are presented, and does not require the approach to be re-trained on the entire previous dataset. In this paper, we propose Incremental VSEARCH (I-VSEARCH) and its semi-supervised version for taxonomic classification, as well as a threshold independent VSEARCH (TI-VSEARCH) as wrappers around VSEARCH, a well-established (unsupervised) clustering algorithm for metagenomics. We show - on a 16S rRNA gene dataset - that I-VSEARCH, running incrementally only on the new batches of data that become available over time, does not lose any accuracy over VSEARCH that runs on the full data, while providing attractive computational benefits.