Engineering cellular phenotypes often requires the regulation of many genes. When using CRISPR interference, co-expressing many single-guide RNAs (sgRNAs) triggers genetic instability and phenotype loss, due to the presence of repetitive DNA sequences. We stably co-expressed 22 sgRNAs within non-repetitive extra-long sgRNA arrays (ELSAs) to simultaneously repress up to 13 genes by up to 3500-fold. We apply biophysical modeling, biochemical characterization, and machine learning to develop toolboxes of non-repetitive genetic parts, including 28 sgRNA handles that bind Cas9. We design ELSAs by combining non-repetitive genetic parts according to algorithmic rules quantifying DNA synthesis complexity, sgRNA expression, sgRNA targeting, and genetic stability. Using ELSAs, we created three highly selective phenotypes in Escherichia coli, including redirecting metabolism to increase succinic acid production by 150-fold, knocking down amino acid biosynthesis to create a multi-auxotrophic strain, and repressing stress responses to reduce persister cell formation by 21-fold. ELSAs enable simultaneous and stable regulation of many genes for metabolic engineering and synthetic biology applications.