Yes, you can correlate anything to anything, but a correlation will not reveal the underlying mechanism at play. I would not worry too much about the differences in scale between the expression data and the methylation data. Just ensure that the expression datasets are normalised in the same way. If using RSEM, it may help to log these in order to bring the distribution to a normal distribution.

There are some previous threads:

• What is a typical workflow to correlate methylation and expression data?
• Correlation between methylation (450K) and gene expression (RNA-Seq)
• Correlation between gene expression and methylation
• scRNA-seq and Chip-seq data integration

Slightly related but similar idea:

• GLM to predict what SNP influences DNA-methylation

Another possibility, which is preferable, is to build linear regression models between each gene's expression and the corresponding methylation probes mapping to it (one-to-many). From these models, you can easily derive both a p-value and a r-squared value, along with many other things.