Use bedtools subtract to get gff of interest minus gff repeats, then use featureCounts or bedtools coverage using the resulting gff to count reads mapping to the remaining features.

Convert to BED via convert2bed helper scripts:

$ gff2bed < annotations.gff > annotations.bed
$ gff2bed < rmsk.gff > rmsk.bed
$ bam2bed < reads.bam > reads.bed

If you want at least X reads that overlap annotations that do not overlap repeatmasked regions:

$ X=1234
$ bedmap --count --echo --delim '\t' annotations.bed reads.bed | awk -vX=${X} '$1 >= X' | cut -f2- | bedops -n 1 - rmsk.bed > answer.bed

(Replace X=1234 with whatever threshold you want.)

The file answer will contain annotations that meet your read threshold and which do not overlap repeatmasked regions.

You could instead do conversion, mapping, and filtering with the following one-liner, which avoids making intermediate files and so will be even faster than the usual BEDOPS speedup:

$ gff2bed < annotations.gff | bedmap --count --echo --delim '\t' - <(bam2bed < reads.bam) | awk -vX=${X} '$1 >= X' | cut -f2- | bedops -n 1 - <(gff2bed < rmsk.gff) > answer.bed