You can have a DWORD resultant from a math instruction that overflows the 32767 maximum value for an Integer. When a math instruction overflows, the unsigned remainder is put into the regular destination address and the 32-bit result is placed into a special "Math Register" comprised of processor status words S:13 and S:14.
The data is only there until some other instruction uses the math register, so immediately after the math instruction that results in overflow, you should COPY these out into 16-bit holding registers to get ready to put them into the DeviceNet I/O or message area.
I've done this to set the Index position (400,000 counts or so) for an A-B Ultra3000i motion controller on DeviceNet and it works fine. In that case, I didn't have to reverse the order of the Words, but for your device you might, depending on what sequence it expects the data bytes on DeviceNet.