I was an embedded developer for years for critical applications that could not go down. While I preferred avoiding dynamically allocating memory, as it was much less risky, there were certainly times it just made sense or was the only way.
One was when we were reprogramming the device, which was connected to an fpga which would also require reprogramming. You couldn’t store both the fpga binary and the new binary for the device in memory at once, but there was plenty of space to hold each one individually. So allocate the space for the fpga, program it, free and allocate space the new processor code, verify and flash.