I think there are many approaches you can follow. This also depends from the dataset you are working with.

The first thing would be to take a reference protein sequence and use it as query for a blast search against your dataset/genome.

You can then take the hits and use them as query for a second round fo blast search in you dataset. However, this second step might led to a high number of false positive so you may want to check the hits manually.

A second approach is to take a representative number of sequences for the protein you are interested in for public db. You align them, create a Hidden Markov model and use this model to scan your dataset.

I would recommend you to always double check manually your hits.