You can use "bedtools maskfasta". You can get the regions from the UCSC table browser.

I use subtractBed. This helps me retain the bed file architecture with my reads and I can insert it anytime/anywhere in my pipeline after mapping. maskfasta give a fasta output.

Assuming hg19, grab repeat data, if needed:

$ mysql -h genome-mysql.cse.ucsc.edu -A -u genome -D hg19 -e 'select genoName, genoStart, genoEnd, repName, swScore, strand from rmsk' | tail -n +2 > repeats.bed
$ head -3 repeats.bed
chr1    10000    10468    (CCCTAA)n    1504    +
chr1    10468    11447    TAR1    3612    -
chr1    11503    11675    L1MC    437    -

Then perform set operations with BEDOPS:

$ bedops -n 1 <(bam2bed < reads.bam) repeats.bed > reads_that_do_not_overlap_repeats.bed

Then convert the result to the desired end format.