Genomic sequences contain instructions for protein and cell production. Therefore understanding and dentification of biologically and functionally meaningful patterns in DNA sequences is of paramount importance. Modeling of DNA sequences in its turn can help to better understand and identify such patterns and dependencies between them. It is well-known that genomic data contains various regions with distinct functionality and thus also statistical properties. In this work we focus on modeling of such individual regions of distinct functionalities. We apply the concept of context trees to model these DNA regions. Based on the Minimum Description Length principle, we use the estimated compression rate of a genomic region, given such models, as a similarity measure. We show that the constructed model can be used to distinguish specific genes within DNA sequences.