Can't you just use the sample() function? Something like:

## assuming you have a vector of all the genes called "allGenesName"
n <- 1000
p <- 114
rand114Genes <- matrix(nrow=n, ncol=p)

for (i in 1:n) {
rand114Genes[i,] <- sample(allGenesName, p)
}
## and then you do your previous analysis in a loop (or using "apply()") for all lines in rand114Genes
## at last you compile your classification performances

Just an add-on remark: to me, 114 genes seem a lot for a signature set. If you have not done it yet, I would suggest you to try sorting them by weight in the classification process, and try iteratively from 1 to 114 to see how the classification performance evolves (it may be possible that the ~15 first give you the same result).

At last, have you considered any cross-validation technique in your meta-analysis ?