So if I get you right, you want traffic from a CPE somewhere deep in your network to be handled in a specific way on some border router. But the choice of this handling should be made already at the CPE itself and you are looking for a way to convey the information about the choice to the border element across the network path. For some reason, you cannot use the WAN IP address of the CPE as a basis to choose the handling on the border router itself (although you haven't mentioned any NAT on the intermediate elements on the path between the CPE and the border router).
The most straightforward, and most ugly, way is to misuse the DSCP field of the iP header to convey the information. In this case, you would use the mangle rules on the CPE to set the DSCP field of the packet rather than to assign a routing-mark to it.
Anything else requires some kind of tunneling. You can use a point-to-point tunnel from each such CPE to the border element (which means you'll have a dedicated virtual interface for each CPE at the border element) or, if your network is or can be made L2-transparent, you may use a network-wide VLAN where the CPEs would attach a /32 IP address to the correspoding /interface vlan and set its network parameter to an IP address which would be put up on the corresponding /interface vlan on the border element. In either case, the return path (from the border router to the CPE) can be the basic one, but in such case you must not use any restrictive rules in stateful firewalls on the intermediate elements (as these would see only one direction of each connection and block most traffic as invalid). With PtP tunnels, you could use the tunnels also as the backward routes to the CPEs but it would be a configuration nightmare.
With the VLAN approach you might be able to use the VLAN also for response traffic if you set the IP address at the border element side with a mask spanning all the CPE WAN addresses, put the VLAN interface into a VRF, and set arp=local-proxy-arp at the CPE's /interface vlan. This way, you could make the border router use the VLAN as a return path (using connection-mark and routing-mark matching the VRF one), as the CPEs would respond to the ARP requests received at the VLAN interface regarding an IP address assigned to another interface. But that's a speculation, I have never tested anything this wild.
Instead of writing novels, post /export hide-sensitive. Use find&replace in your favourite text editor to systematically replace all occurrences of each public IP address potentially identifying you by a distinctive pattern such as my.public.ip.1.