The numbers you are quoting are MASK numbers. These numbers work with an IP address to determine which binary bits of the IP number are the network number, and which are the node number on that network. The mask 255.255.255.0 says that the first 24bits are network, the last 8 are node. 255.255.255.224 says the first 27 bits are network, the last 5 are node.
If your real IP number ranges are, for example, 192.168.0.0 for your local network, 192.168.1.0 for the WAN, and 10.12.14.32 for the ISP, you probably have a layout like this:
ISP--routerA--10.12.14.32--routerB--192.168.0.0--routerC--192.168.2.0
There may actually be a link and a routerD off to the right of routerC, but we can ignore that for this example.
RouterA is probably supplied by the ISP, and you may not have programming access to it. If you do not, you may need to ask your ISP to make a change in it for you.
Your workstations are, on this drawing, all in subnet 192.168.0.0. The easiest way to set up everything is to have all of them point to RouterB as their default gateway. RouterB should point to RouterA as its default gateway and have static routes pointing to RouterC for any network on the far side of RouterC. RouterB should provide NAT services for all addresses in 192.168.0.0 for traffic to the internet.
RouterA should have the ISP as its default gateway, and have static routes to RouterB for all of your internal networks. All machines in the 10.12.14.32 subnet should have routerA as their default gateway.
If you cannot program RouterA, and your ISP will not do it for you, make RouterB the default gateway for all machines in 10.12.14.32. (It will mean one extra hop for internet traffic originating in 10.12.14.32, but it will at least work.)
Good luck.