Those journals that ask for unprocessed original images of gels and western blots often publish the data as supplementary materials. In such cases, you do not need to keep the data at all, as the published article acts as a sort of repository, where your data will be stored in perpetuity.
Other journals may ask you to upload the unprocessed data to a reputable repository, and a link to the images will be provided in the published article. You should consider uploading all original data to a repository even if your target journal does not ask for the unprocessed western blot images. It is a simple, convenient way to archive data from your lab.
If you wish to retain the data within your lab, check with your funder and/or university for specific instructions. For example, University College London (UCL) specifies that data should be retained for 10 years from the date of submission, whereas the National Institutes of Health (NIH) specifies 3 years. If your funder/university does not have a specific data retention policy, you should err on the side of caution by retaining the data for 10 years. Many papers are retracted each year because the authors are unable to provide the original data.