GMS can be transferred to a new line in a similar way as we transfer any other traits governed by recessive allele. Cross the parent 1 (MsMs) as male parent to which we want to transfer GMS with male sterile parent 2 carrying homozygous recessive alleles (msms). Cross the resultant F1 (Msms) with recurrent parent 1 (MsMs) and grow BC1 plants and self each plant and collect BC1F2 seed. In F2, identify the sterile plants (msms) and cross with recurrent parent 1 (MsMs) to get BC2F1 plants and then cross BC2F1 plants again with recurrent parent 1 (MsMs) to get BC3F1 plants. Self the BC3F1 plants to get BC3F2 plants and identify the sterile plants. Repeat this procedure up to BC6 or BC7 generation to recover sterile plants having more than 99% genome of recurrent parent 1. Finally we can get a new GMS line which can used in hybrid seed production program. The new GMS line designated as 'line A' is to me maintained by crossing it with original 'line B', the recurrent parent. The F1 always carries about 50% fertile and 50% sterile plants and the fertile plants are to be removed from male sterile line A before pollen shedding for true hybrid seed production.
Only suggestion to improve the above description is that you need to self at least 5 BC1F1 plants to have a 95% probability of obtaining at least one BC1F2 population segregating for the recessively inherited genetic male-sterility, assuming that a single genetic locus is involved in its control, and should evaluate at least 11 plants in each BC1F2 to have a similar probability of phenotypically detecting segregation for such a single recessive allele. There is a nice, short Crop Science paper from 1977 (?) that provides tables for estimating the numbers of plants required to detect particular segregation patterns -- also related tables in the Appendices of the statistics text by Steele & Torrie.