You may try to experiment with fake quality values for these sequences which miss them. One may start with just putting 4s, 10s or anything in between (http://www.phrap.com/phred/) and check if there are big differences in assembly.

More accurate approach (assuming the same sequencing technology/chemistry) would be to calculate average base quality at a given position from reads you have real values, and put these for reads without quality values.

You may also assembly reads with qualities first, then map the remaining ones to existing contigs. Use the mapping quality to estimate quality of the reads (lower the quality for mismatches), but this is probably an overkill.