When big data actually are low-rank, or entrywise approximation of certain function-generated matrices